9:15 AM-10:00 AM
Room: Rio Mar 5
Chair: Richard A. Tapia, Rice University, USA
Over the last several years, my research team at the University of Minnesota's Laboratory for Computational Science & Engineering (LCSE) has been exploiting clusters of powerful shared memory multiprocessors (SMPs) containing up to 3800 processors to simulate turbulent compressible fluid flows on grids of up to 8 billion cells. The unprecedented computational speed of these SMP cluster systems has allowed us to simulate not only the principal energy-containing scales of complex fluid flows but also to capture as well the turbulent motions on finer scales which these flows induce. These new simulations give us a new level of confidence in our results. They are for the first time accurate enough to be used to help guide efforts to construct theoretical models of the statistically averaged behavior of the turbulence in flows of this type. Simulations of compressible convection in stars, of turbulent mixing of fluids at shock-accelerated interfaces, and of homogeneous, compressible turbulence itself will be presented. These turbulent flow visualizations have been constructed at the LCSE by our PowerWall visualization system from multiple terabytes of archived data from these simulations. The parallel implementation of the PPM gas dynamics code which carried out these calculations on large SMP clusters will be described along with numerical algorithm modifications targeted for these clusters. In addition, the visualization techniques developed at the LCSE, which make it possible to view the data interactively, will be described.
Paul R. Woodward